Goldman Sachs restricts AI coding tools to engineers, locking out traders and bankers after internal review
Goldman Sachs restricts AI coding tools to engineers, locking out traders and bankers after internal review #
Goldman Sachs has pulled access to advanced AI coding tools from most of its front-office workforce, limiting Anthropic’s Claude Code and Microsoft’s GitHub Copilot to the bank’s software engineers after an internal review of productivity and risk.
Updated internal guidance issued in October separated Goldman’s roughly 12,000 professional software engineers from the traders and investment bankers who had spent months using the tools in their daily work. Bloomberg reported that some traders had built the software into core workflows, including monitoring client trading flows, simulating shifts in market positions, and producing written and graphical materials. The change has frustrated affected staff, sources familiar with the matter told Bloomberg.
Goldman spokesperson Nick Carcaterra confirmed the shift in a statement, saying the bank makes certain AI tools available to employees to identify the best use cases across different parts of the firm. “After a period of experimentation where we assess a variety of factors including productivity and risk management, we put them in the hands of those who are best suited to leverage them for the long term,” Carcaterra said.
The policy exempts Goldman’s so-called “strats,” a specialist group whose roles combine software engineering, quantitative research, and trading, who retain access to the restricted tools. For mainstream traders and bankers, the reversal contradicts the position Goldman was publicly promoting as recently as July, when Chief Information Officer Marco Argenti praised the rise of “vibe coding” among non-technical employees. Argenti described front-office staff generating prototype solutions with AI tools and passing the results to engineers for further development.
On trading desks, data barriers, regulatory compliance, and the handling of material non-public information make automated code generation harder to manage. AI-generated code running under a user’s permissions raises questions about auditability, reliability, and information security that are harder to resolve in a regulated environment.
Goldman continues to invest in AI, maintaining a proprietary internal assistant that draws on models from OpenAI, Alphabet, and Anthropic. The bank has not indicated that a specific security incident or system failure prompted the restriction. The decision carries some irony: Goldman is serving as lead underwriter for Anthropic’s planned initial public offering, the same company whose Claude Code product it has now pulled from its trading floors.
Copy link